CombiNAber Explanation
Introduction
We have developed two models, the additive model (Kong et al. 2015) and the Bliss-Hill model (Wagh et al. 2016), that can accurately predict neutralization by antibody combinations using IC_{50} and/or IC_{80} data for individual antibodies tested in vitro against a virus panel. The additive model can work with either IC_{50} or IC_{80} alone, while the Bliss-Hill model requires both. We have found that while both models are fairly accurate in predicting experimental combination data, the Bliss-Hill model is more accurate, and also allows calculation of other neutralization metrics (see Incomplete neutralization).
- IC_{50}/IC_{80} data
The titers are assumed to be in μg/ml, and should be in the following format (we allow tab, comma and space as delimiter):
VirusName mab1:IC50 mab1:IC80 mab2:IC50 mab2:IC80
Vir1 10.0 15.0 20.0 GT25.0
Vir2 NT NT LT0.001 0.02
The first row should contain the entries “Virus name”, followed by IC_{50} and/or IC_{80} data for each mAb, e.g., “mab1:IC50, mab2:IC50,…” for
IC_{50} data only or “mab1:IC50, mab1:IC80, mab2:IC50, mab2:IC80,…” for both IC_{50} & IC_{80} data. “NT” should be used for viruses that were not tested against some antibodies.
For IC_{50}/IC_{80}
titers beyond the experimental ranges, please use “LT” or “GT” followed by titer values, instead of “<” or “>” (which we cannot allow due to cyber-security issues). If there are
“<” or “>” signs in your data, the following error will be generated:
Your input doesn't seem to be in an accepted format...
- Antibody class
We require the user to input the class of each antibody in the data, which is based on the antibody epitope. We require this data in the following format (we allow tab, comma and space as delimiter):
PGT128 V3g
2G12 V3g
new_mab X
In this example, PGT128 and 2G12 are grouped in the same class “V3g”
(V3 glycan). The epitopes for PGT128 & 2G12 are close but slightly different, however, they target the same key Env positions (in this case glycan at 332). For
antibodies targeting Env regions that are different than the common known epitopes and that are are yet to be characterized, we allow a wildcard class (“new_mab” in
this example).
This information is used in two ways. First, if the antibodies target similar Env regions, there can be steric competition between to bind Envs, and for Bliss-Hill model, this
translates into using different equations than the case when antibodies target different regions. Second, antibodies targeting similar Env regions have similar
neutralization profiles (see e.g. Wagh et al. 2016), and hence, it is likely that the virus can evolve
simultaneous resistance to all such antibodies using the same escape mutations. Thus, when looking at coverage by multiple antibodies in the combination (see
Active coverage by multiple antibodies in combination), we consider only antibodies in the combination that target different epitopes (i.e., have different classes).
Analysis Options
Output Data & Analyses
All of the data, figures & analyses generated are available for the user to download as a zip file. Select analyses & figures will be shown on the output page.
- Output page
The output page shows summary tables and figures for best-in-class single mAbs, all mAb combinations with the same number of mAbs, best combinations with different number of mAbs and combinations of interest. The summary tables and description of figures are described below. For figures, IC_{80} breadth-potency and
IC_{80} active coverage figures are preferably shown, unless IC_{80} data was not given, in which case IC_{50} figures are shown.
- Zip file of full results
The first level of directory structure in the zip file has directories for each target concentration. Within each target
concentration directory, will be up to 6 directories: “single_mabs”, “2mab_combinations”, “3mab_combinations”, “4mab_combinations”,
“best_combos_diff_no_of_mabs_comparison” (comparison of best combinations with different number of mAbs) and “combinations_of_interest”. Each of the first 4
directories has three sub-directories – “data”, “summary” and “figures”. The last two directories will have “summary” and “figures” (since the data for these combinations
will be in the first 4 directories). The “data” directory has all the raw data of predictions, and the “figures” directory will have all the analysis figures
generated. In the “summary” directory, we will generate summary tables that summarize the different neutralization metrics and offer ranks for the antibody
combinations (see below for ranking explanation). For the first 4 of 6 directories above, two summary tables are generated – one for comparing all analyzed combinations with the same number of mAbs, and one for comparing the best combinations from each class of antibody combinations (defined using the classes of each
antibody in the combination). For the last 2 of 6 directories above, one summary file in each case is generated, either comparing best mAb to best 2mAb, 3mAb & 4mAb
combinations, or comparing all the combinations of interest.
- Ranking of combinations
This tool calculates the rankings for antibody combinations using all of the neutralization metrics chosen by the user.
The rankings are calculated using the summary characteristics shown in the summary tables. The summary characteristics for the neutralization metric are as follows:
- For combination IC_{50}/IC_{80}, the summary characteristics used for ranking are geometric mean titers (using
neutralized viruses) and fraction of the virus panel neutralized (coverage) with IC_{50}/IC_{80} less than the target concentration(s). If available,
IC_{80} titers are preferred over IC_{50} titers (so, IC_{50} titers are used only if IC_{80} titers are not available).
- For coverage with multiple mAbs active, the summary characteristics used are geometric mean combination IC_{50}/IC_{80} titers and coverage of the
virus panel using only those viruses that are actively neutralized by multiple mAbs in the combination, with user-specified number of active mAbs and with user-specified target
concentration(s) as the mAb activity threshold. If available, IC_{80} titers are preferred over IC_{50} titers (so, IC_{50} titers are used only if
IC_{80} titers are not available).
- For incomplete neutralization, the summary characteristics used are median incomplete neutralization values and the fraction of viruses neutralized to
more than 95% level (>0.95) by antibodies/antibody combinations at the target concentration. If both incomplete neutralization and IIP are chosen, then only the
fraction of viruses neutralized to more than 95% levels is used. (This is because the IIP and incomplete neutralization for our predictions are related
mathematically, and the medians of both these metrics are perfectly correlated).
- For instantaneous inhibitory potential (IIP), the summary characteristics used are median IIP values and
the fraction of viruses with IIP > 5. The IIP > 5 threshold was found to be correlated with efficacy of combination antiretroviral therapy
(Jilek et al. 2012).
The ranking of combinations is done in two steps. First, for each combination, each of the available above summary
characteristics is ranked when compared with all other combinations with the same number of mAbs. The appropriate ordering of characteristics is taken into account
(e.g. lower geometric means are better, higher coverage values are better), and tied values for characteristics get the same rank, without advancing the ranks
accounting for tied values (e.g. if two combinations using geometric mean IC_{80} are ranked 1, then the combination with next best geometric mean
IC_{80} gets rank 2). Next, each combination is ranked based on the average of the ranks for all of its characteristics. In this ranking scheme, tied combinations
get the same rank, however, ranks are advanced to account for tied combinations with better rank (e.g. if two combinations have ranks 1, then the next best
combination will have rank of 3). These ranks are found in the summary tables for all combinations analyzed.
The ranks for each combination are used to find the best-in-class combinations. These combinations are reported in the best-in-class summary tables, and
their data is shown/highlighted in the analysis figures. If there are ties for best-in-class combinations, two additional methods are used to try and break ties.
First method to break ties uses only combinations belonging to the same class (as opposed to the above method, which uses all combinations with same number of mAbs)
to rank the summary characteristics of the tied combinations, and then, uses the average rank of these characteristics to rank the combinations. If the best
combinations still have same average ranks, then sum of squared ranks for each summary characteristic is used to break the ties. The rationale behind using sum of
squared ranks is that it chooses combinations that have better ranks overall, by more severely penalizing lower ranks. For example, if one combination has ranks
(1,4) for two characteristics and another combination has ranks (2,3), then the latter combination will be chosen by sum of squared ranks criterion. If both of
these methods fail to resolve the ties, then one combination is picked arbitrarily as the best-in-class combination, and a note is added to the summary table for the
best-in-class combinations indicating the combinations were tied.
When ranking combinations of interest, the rank of each combination within the full set of combinations with the same number of mAbs is reported. However,
they are also ordered based on the comparison within the set of combinations of interest alone, using the baseline algorithm above, with those characteristics that are
applicable to all combinations of interest (e.g. coverage with 3 mAbs active is not applicable to 2mAb combinations of interest).
- Figures
Overall breadth-potency:
These figures show the cumulative coverage of the virus panel (a fraction between 0 & 1) at various IC_{50}/IC_{80} values for single mAbs and/or mAb combinations.
Active Coverage with at least “n” mAbs active:
These figures show cumulative coverage of the virus panel (a fraction between 0 & 1) at various combination IC_{50}/IC_{80} by considering only those viruses that were also actively neutralized by at least “n” of the mAbs in the combination using single IC_{50} or IC_{80} thresholds of target concentration values.
Incomplete Neutralization:
These figures show the neutralized fraction for each virus in the panel by single mAbs/mAb combinations at the target concentration values. The medians (thick black lines) and quartiles
Instantaneous Inhibitory Potential (IIP):
These figures show the IIP values for each virus in the panel for single mAbs/mAb combinations at the target concentration values. The medians (thick black lines) and quartiles (thin black lines) are also show for each single mAb or mAb combinations.
last modified: Wed Oct 12 11:58 2016